Speech recognition method, device and terminal

Abstract

The present invention provides a speech recognition method, device and terminal to solve the problems of high error rate and poor domain compatibility in speech recognition using existing speech recognition technology. The method of the present invention comprises: obtaining speech data to be recognized, the speech data to be recognized including speech data from multiple fields; performing speech recognition on the speech data to be recognized based on a pre-constructed first speech recognition model. The present invention obtains speech data to be recognized, the speech data to be recognized including speech data from multiple fields; performing speech recognition on the speech data to be recognized based on a pre-constructed first speech recognition model. In this way, by using the first speech recognition model with domain compatibility, words in different fields can be accurately recognized, thereby improving the accuracy of speech recognition.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a speech recognition method, device and terminal. Background Art

[0002] With the development of artificial intelligence, speech recognition has made tremendous progress. Existing technologies primarily target neural networks, while other network structures, particularly algorithms without gradient descent, cannot be used. Alternatively, meta-learning methods are used to automatically estimate subtask weights. However, meta-learning aims to learn a generalized model that performs well on new tasks, which may not necessarily be optimal for the current task. Alternatively, these approaches only consider static sample annotations, ignoring the recognition performance across two domains during each iteration. This results in high error rates and poor domain compatibility when using existing speech recognition technologies. Summary of the Invention

[0003] The object of the present invention is to provide a speech recognition method, device and terminal for solving the problems of high error rate and poor domain compatibility in speech recognition using existing speech recognition technology.

[0004] In order to achieve the above object, the present invention provides a speech recognition method, comprising:

[0005] Acquiring voice data to be recognized, wherein the voice data to be recognized includes voice data in multiple fields;

[0006] Based on a pre-built first speech recognition model, speech recognition is performed on the speech data to be recognized.

[0007] Before obtaining the voice data to be recognized, the method further includes:

[0008] Acquire first corpus data, where the first corpus data includes speech data from multiple fields;

[0009] Performing model training based on the first corpus data to obtain a second speech recognition model;

[0010] Verifying the second speech recognition model to obtain a ratio of missed and false positives for target domain words;

[0011] According to the missed detection and false detection ratio, the second speech recognition model is adjusted to obtain the first speech recognition model.

[0012] The missed detection of the target domain words includes: the number of times the target domain words are recognized as words from other fields, and the number of times the target domain words are not recognized;

[0013] The misidentification of the target domain words includes: the number of times other domain words are recognized as the target domain words, and the number of times they are redundantly recognized as the target domain words.

[0014] The verifying of the first speech recognition model to obtain the ratio of missed and false positives of target domain words includes:

[0015] Performing verification and recognition on the second speech recognition model using verification data to obtain a recognition result, wherein the verification data is part of the first corpus data;

[0016] According to the recognition results, the ratio of missed detection to false detection of words in the target domain is obtained.

[0017] The step of obtaining the ratio of missed searches to false searches of target domain words based on the recognition results includes:

[0018] Counting the recognition results to obtain missed and misidentified words in the target domain;

[0019] According to the missed searches and mis-searches of the target domain words, a missed search and mis-search ratio of the target domain words is calculated.

[0020] Before verifying the second speech recognition model and obtaining the ratio of missed detection to false detection of target domain words, the method further includes:

[0021] Performing word segmentation on the text data corresponding to the first corpus data to obtain a plurality of words;

[0022] The multiple words are screened based on the fields to which they belong to obtain field words corresponding to the respective fields.

[0023] The adjusting the first speech recognition model according to the missed detection to false detection ratio includes:

[0024] When the ratio of missed detection to false detection is greater than 1, increasing the weight coefficient of the sample loss from the target domain in the loss function;

[0025] When the ratio of missed detection to false detection is less than 1, the weight coefficient of the sample loss from the target field in the loss function is reduced, and the loss function is the weighted sum of the sample losses in multiple fields.

[0026] The present invention also provides a speech recognition device, comprising:

[0027] A first acquisition module is used to acquire voice data to be recognized, wherein the voice data to be recognized includes voice data in multiple fields;

[0028] The speech recognition module is used to perform speech recognition on the speech data to be recognized based on a pre-built first speech recognition model.

[0029] The present invention further provides a terminal, comprising a processor and a transceiver, wherein the transceiver receives and sends data under the control of the processor, and the processor is configured to perform the following operations:

[0030] Acquiring voice data to be recognized, wherein the voice data to be recognized includes voice data in multiple fields;

[0031] Based on a pre-built first speech recognition model, speech recognition is performed on the speech data to be recognized.

[0032] The processor is further configured to:

[0033] Acquire first corpus data, where the first corpus data includes speech data from multiple fields;

[0034] Performing model training based on the first corpus data to obtain a second speech recognition model;

[0035] The second speech recognition model is verified to obtain the ratio of missed and false positives of target domain words.

[0036] According to the missed detection and false detection ratio, the second speech recognition model is adjusted to obtain the first speech recognition model.

[0037] The missed detection of the target domain words includes: the number of times the target domain words are recognized as words from other fields, and the number of times the target domain words are not recognized;

[0038] The misidentification of the target domain words includes: the number of times other domain words are recognized as the target domain words, and the number of times they are redundantly recognized as the target domain words.

[0039] The processor is further configured to:

[0040] Performing verification and recognition on the second speech recognition model using verification data to obtain a recognition result, wherein the verification data is part of the first corpus data;

[0041] According to the recognition results, the ratio of missed detection to false detection of words in the target domain is obtained.

[0042] The processor is further configured to:

[0043] Counting the recognition results to obtain missed and misidentified words in the target domain;

[0044] According to the missed searches and mis-searches of the target domain words, a missed search and mis-search ratio of the target domain words is calculated.

[0045] The processor is further configured to:

[0046] Performing word segmentation on the text data corresponding to the first corpus data to obtain a plurality of words;

[0047] The multiple words are screened based on the fields to which they belong to obtain field words corresponding to the respective fields.

[0048] The processor is further configured to:

[0049] When the ratio of missed detection to false detection is greater than 1, increasing the weight coefficient of the sample loss from the target domain in the loss function;

[0050] When the ratio of missed detection to false detection is less than 1, the weight coefficient of the sample loss from the target field in the loss function is reduced, and the loss function is the weighted sum of the sample losses in multiple fields.

[0051] The present invention also provides a terminal, comprising a memory, a processor, and a program stored in the memory and executable on the processor; when the processor executes the program, the speech recognition method as described above is implemented.

[0052] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which implements the steps in the above-mentioned speech recognition method when executed by a processor.

[0053] The above technical solution of the present invention has at least the following beneficial effects:

[0054] In an embodiment of the present invention, voice data to be recognized is obtained, and the voice data to be recognized includes voice data from multiple fields; based on a pre-built first voice recognition model, voice recognition is performed on the voice data to be recognized. In this way, by using the first voice recognition model with field compatibility, words from different fields can be accurately recognized, thereby improving the accuracy of voice recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A schematic diagram showing a flow chart of a speech recognition method according to an embodiment of the present invention;

[0056] Figure 2 A schematic diagram showing modules of a speech recognition device according to an embodiment of the present invention;

[0057] Figure 3 A schematic diagram showing the structure of a terminal according to an embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0059] The present invention aims to solve the problems of high error rate and poor domain compatibility in speech recognition using existing speech recognition technology, and provides a speech recognition method and terminal.

[0060] In order to enable those skilled in the art to better understand the method of the embodiment of the present invention, the following description is first given.

[0061] For the binary classification problem, define the following four quantities:

[0062] True positives (TP): The number of examples where the standard answer is positive and the system classification is also positive;

[0063] False negatives (FN): The number of examples where the standard answer is positive but the system classifies them as negative;

[0064] False positives (FP): The number of examples where the standard answer is negative but the system classifies them as positive;

[0065] True negatives (TN): The number of examples where the standard answer is negative and the system classification is also negative;

[0066] And record the total number of positive and negative examples as:

[0067] n + =TP+FN (1)

[0068] n - =TN+FP (2)

[0069] Recall rate of positive class: R + =TP / (TP+FN)=TP / n + (3)

[0070] Recall rate of negative class: R - =TN / (TN+FP)=TN / n - (4)

[0071] Positive class omission rate: 1-R + =1-TP / (TP+FN)=FN / n + (5)

[0072] Error rate of negative class: 1-R - =1-TN / (TN+FP)=FP / n - (6)

[0073] like Figure 1FIG. 1 is a flow chart of a speech recognition method according to an embodiment of the present invention, which specifically includes:

[0074] Step 101: Acquire speech data to be recognized, wherein the speech data to be recognized includes speech data in multiple fields;

[0075] In this step, the voice data in multiple fields may be voice data in multiple fields such as express delivery, finance, education, insurance, and real estate.

[0076] Among them, the voice data to be recognized can be obtained by answering high-frequency calls (such as harassing calls).

[0077] Step 102: Perform speech recognition on the speech data to be recognized based on a pre-built first speech recognition model.

[0078] Here, the first speech recognition model has domain compatibility and can recognize speech data in different domains.

[0079] The speech recognition method of an embodiment of the present invention obtains speech data to be recognized, which includes speech data from multiple fields; and performs speech recognition on the speech data to be recognized based on a pre-built first speech recognition model. In this way, by using the first speech recognition model with field compatibility, words from different fields can be accurately recognized, thereby improving the accuracy of speech recognition.

[0080] As an optional implementation, before step 101, the method of the embodiment of the present invention may further include:

[0081] Acquire first corpus data, where the first corpus data includes speech data from multiple fields;

[0082] Performing model training based on the first corpus data to obtain a second speech recognition model;

[0083] It should be noted that the first corpus data includes speech data from multiple fields, where "multiple" specifically refers to greater than or equal to two. The first corpus data is a random mixture of speech data from multiple fields.

[0084] Specifically, the first corpus data can be divided into a training set and a validation set, and the training set is used to perform model training to obtain the second speech recognition model.

[0085] Verifying the second speech recognition model to obtain a ratio of missed and false positives for target domain words;

[0086] Here, the target field in the target field term is any one of a plurality of fields.

[0087] Optionally, the missed detection of the target domain words includes: the number of times the target domain words are identified as words from other fields, and the number of times the target domain words are not identified;

[0088] The misidentification of the target domain words includes: the number of times other domain words are recognized as the target domain words, and the number of times they are redundantly recognized as the target domain words.

[0089] It should be noted that the other fields in the term "other fields" refer to other fields among multiple fields except the target field.

[0090] Here, both missed searches and mis-searches of target domain words are errors in identifying target domain words.

[0091] Among them, the target domain word is recognized as a word in other fields and the other field words are recognized as the target domain word, which are both substitution errors, that is, word x is recognized as word y.

[0092] The failure to recognize the target domain word is a deletion error, that is, the word at that position is not recognized, i.e., missed recognition; being redundantly recognized as a target domain word is an insertion error, that is, one more word is recognized, i.e., misrecognition.

[0093] As an optional implementation, the step of verifying the second speech recognition model to obtain the ratio of missed detection to false detection of target domain words may specifically include:

[0094] Performing verification and recognition on the second speech recognition model using verification data to obtain a recognition result, wherein the verification data is part of the first corpus data;

[0095] Here, the verification data may be data in a verification set divided from the first corpus data.

[0096] It should be noted that, during the model training process, after each round of iteration or a preset number of iterations, the second speech recognition model is verified and recognized using verification data to obtain a recognition result.

[0097] According to the recognition results, the ratio of missed detection to false detection of words in the target domain is obtained.

[0098] Here, this step may specifically include:

[0099] Counting the recognition results to obtain missed and misidentified words in the target domain;

[0100] For example, multiple fields include field A and field B, where when the target field word is a field A word, the recognition results are counted to obtain missed detections of field A words, that is, the number of times field A words are recognized as field B words and the number of times field A is not recognized are counted; and the recognition results are counted to obtain false detections of field A words, that is, the number of times field B words are recognized as field A words and the number of times they are redundantly recognized as field A words are counted.

[0101] The target domain word is domain B. Please refer to the above situation and will not repeat it here.

[0102] According to the missed searches and mis-searches of the target domain words, a missed search and mis-search ratio of the target domain words is calculated.

[0103] Here, in order to obtain the omission and false detection rates of target domain words and facilitate the subsequent calculation of the omission and false detection ratio of the target domain words, as an optional implementation, before verifying the second speech recognition model and obtaining the omission and false detection ratio of the target domain words, the method of the embodiment of the present invention may further include:

[0104] Performing word segmentation on the text data corresponding to the first corpus data to obtain a plurality of words;

[0105] The multiple words are screened based on the fields to which they belong to obtain field words corresponding to the respective fields.

[0106] This step may specifically include: performing word screening on the multiple words based on the fields to which they belong, that is, removing words that exist in any two or any multiple fields in the multiple fields, and using the remaining words as the field words of the respective fields.

[0107] According to the missed detection and false detection ratio, the second speech recognition model is adjusted to obtain the first speech recognition model.

[0108] This step may specifically include:

[0109] When the ratio of missed detection to false detection is greater than 1, increasing the weight coefficient of the sample loss from the target domain in the loss function;

[0110] When the ratio of missed detection to false detection is less than 1, the weight coefficient of the sample loss from the target field in the loss function is reduced, and the loss function is the weighted sum of the sample losses in multiple fields.

[0111] It should be noted that if the ratio of missed detection to false detection is greater than 1, it means that the missed detection of words in the target domain is greater than the false detection, and it is necessary to increase attention to the target domain, that is, increase the weight coefficient of the sample loss from the target domain in the loss function.

[0112] If the ratio of missed detection to false detection is less than 1, it means that the false detection of words in the target domain is greater than missed detection. It is necessary to reduce the focus on the target domain, that is, to reduce the weight coefficient of the sample loss from the target domain in the loss function.

[0113] Here, the loss function is the weighted sum of losses in different fields:

[0114] Loss_all=∑a_i*Loss_i (7)

[0115] Among them, Loss_all is the total loss of the model, Loss_i is the loss of samples from domain i, and a_i is the weight coefficient of the sample loss of domain i.

[0116] Here, a_i=M_i / F_i (8)

[0117] Among them, Mi_i is the missed detection of field i, and F_i is the false detection of field i.

[0118] When Mi>Fi, the number of missed detections is greater than the number of false detections, and more attention needs to be paid to field i, so its weight coefficient needs to be increased. At this time, ai>1;

[0119] On the contrary, when Mi<F_i, that is, the number of false positives is greater than the number of missed positives, it is necessary to reduce the attention paid to field i, and therefore its weight coefficient needs to be reduced, and at this time a_i<1.

[0120] Based on this, the contribution of each field's corpus to the weight can be adjusted dynamically in real time according to the recognition of words in different fields during the iteration process.

[0121] It should be noted that the first speech recognition model of the embodiment of the present invention focuses on the universality of the model for different fields of the current task, while reducing the confusion between words in different fields. In particular, when the corpus in each field is unbalanced, it suppresses the contribution of strong fields (rich corpus) and improves the contribution of weak fields (sparse corpus). Therefore, in later applications, it can accurately identify speech data in different fields and improve the accuracy of speech recognition. Moreover, the first speech recognition model of the embodiment of the present invention has no restrictions on network structure and learning algorithm.

[0122] The speech recognition method of an embodiment of the present invention obtains speech data to be recognized, which includes speech data from multiple fields; and performs speech recognition on the speech data to be recognized based on a pre-built first speech recognition model. In this way, by using the first speech recognition model with field compatibility, words from different fields can be accurately recognized, thereby improving the accuracy of speech recognition.

[0123] like Figure 2 As shown, an embodiment of the present invention further provides a speech recognition device, the device comprising:

[0124] A first acquisition module 201 is configured to acquire speech data to be recognized, wherein the speech data to be recognized includes speech data in multiple fields;

[0125] The speech recognition module 202 is configured to perform speech recognition on the speech data to be recognized based on a pre-built first speech recognition model.

[0126] Optionally, the speech recognition device further includes:

[0127] A second acquisition module is used to acquire first corpus data, where the first corpus data includes speech data in multiple fields;

[0128] A model training module, configured to perform model training based on the first corpus data to obtain a second speech recognition model;

[0129] A model verification module, configured to verify the second speech recognition model and obtain a ratio of missed and false positives of target domain words;

[0130] The first processing module is used to adjust the second speech recognition model according to the missed detection and false detection ratio to obtain the first speech recognition model.

[0131] Optionally, the missed detection of the target domain words includes: the number of times the target domain words are identified as words from other fields, and the number of times the target domain words are not identified;

[0132] The misidentification of the target domain words includes: the number of times other domain words are recognized as the target domain words, and the number of times they are redundantly recognized as the target domain words.

[0133] Optionally, the model verification module may include:

[0134] a verification unit, configured to verify and recognize the second speech recognition model using verification data to obtain a recognition result, wherein the verification data is part of the first corpus data;

[0135] The first processing unit is configured to obtain a ratio of missed searches to false searches of target domain words according to the recognition result.

[0136] Optionally, the first processing unit is specifically configured to:

[0137] Counting the recognition results to obtain missed and misidentified words in the target domain;

[0138] According to the missed searches and mis-searches of the target domain words, a missed search and mis-search ratio of the target domain words is calculated.

[0139] Optionally, the speech recognition device may further include:

[0140] A second processing module is used to perform word segmentation processing on the text data corresponding to the first corpus data to obtain a plurality of words;

[0141] The word screening module is used to screen the multiple words based on the fields to which they belong, and obtain field words corresponding to each field.

[0142] Optionally, the first processing module includes:

[0143] A second processing unit is configured to increase a weight coefficient of sample loss from the target domain in a loss function when the ratio of missed detection to false detection is greater than 1;

[0144] The third processing unit is used to reduce the weight coefficient of the sample loss from the target field in the loss function when the missed detection and false detection ratio is less than 1, and the loss function is the weighted loss of samples from multiple fields.

[0145] The speech recognition device of an embodiment of the present invention obtains speech data to be recognized, which includes speech data from multiple fields; and performs speech recognition on the speech data to be recognized based on a pre-built first speech recognition model. In this way, by using the first speech recognition model with field compatibility, it is possible to accurately recognize words from different fields and improve the accuracy of speech recognition.

[0146] It should be noted here that the above-mentioned device provided by the embodiment of the present invention can implement all the method steps implemented by the above-mentioned terminal-side method embodiment, and can achieve the same technical effect. The parts and beneficial effects that are the same as the method embodiment in this embodiment will not be described in detail here.

[0147] In order to better achieve the above goals, Figure 3 As shown, an embodiment of the present invention further provides a terminal, including a processor 300 and a transceiver 310. The terminal also includes a user interface 320. The processor 300 is configured to perform the following process:

[0148] Acquiring voice data to be recognized, wherein the voice data to be recognized includes voice data in multiple fields;

[0149] Based on a pre-built first speech recognition model, speech recognition is performed on the speech data to be recognized.

[0150] Optionally, the missed detection of the target domain words includes: the number of times the target domain words are identified as words from other fields, and the number of times the target domain words are not identified;

[0151] The misidentification of the target domain words includes: the number of times other domain words are recognized as the target domain words, and the number of times they are redundantly recognized as the target domain words.

[0152] Optionally, the processor 300 is further configured to:

[0153] Performing verification and recognition on the second speech recognition model using verification data to obtain a recognition result, wherein the verification data is part of the first corpus data;

[0154] According to the recognition results, the ratio of missed detection to false detection of words in the target domain is obtained.

[0155] Optionally, the processor 300 is further configured to:

[0156] Counting the recognition results to obtain missed and misidentified words in the target domain;

[0157] According to the missed searches and mis-searches of the target domain words, a missed search and mis-search ratio of the target domain words is calculated.

[0158] Optionally, the processor 300 is further configured to:

[0159] Performing word segmentation on the text data corresponding to the first corpus data to obtain a plurality of words;

[0160] The multiple words are screened based on the fields to which they belong to obtain field words corresponding to the respective fields.

[0161] Optionally, the processor 300 is further configured to:

[0162] When the ratio of missed detection to false detection is greater than 1, increasing the weight coefficient of the sample loss from the target domain in the loss function;

[0163] When the ratio of missed detection to false detection is less than 1, the weight coefficient of the sample loss from the target field in the loss function is reduced, and the loss function is the weighted sum of the sample losses in multiple fields.

[0164] The terminal of an embodiment of the present invention obtains voice data to be recognized, which includes voice data from multiple fields; and performs voice recognition on the voice data to be recognized based on a pre-built first voice recognition model. In this way, by using the first voice recognition model with field compatibility, it is possible to accurately recognize words from different fields and improve the accuracy of voice recognition.

[0165] An embodiment of the present invention also provides a terminal, including a memory, a processor, and a program stored in the memory and runnable on the processor; when the processor executes the program, the various processes in the speech recognition method embodiment described above are implemented, and the same technical effect can be achieved. To avoid repetition, they will not be described here.

[0166] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When executed by a processor, the program implements the various processes in the above-described speech recognition method embodiment and achieves the same technical effects. To avoid repetition, the details are not described here. The computer-readable storage medium may be, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0167] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0168] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 means for performing functions specified in one or more processes and / or one or more blocks.

[0169] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable storage medium produce a paper product including an instruction device that implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0170] These computer program instructions can also be loaded onto a computer or other programmable data processing device to cause the computer or other programmable device to execute a series of operating steps to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable device to implement the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0171] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A speech recognition method, characterized in that: include: Acquiring voice data to be recognized, wherein the voice data to be recognized includes voice data in multiple fields; Performing speech recognition on the speech data to be recognized based on a pre-built first speech recognition model; Before acquiring the voice data to be recognized, the method further includes: Acquire first corpus data, where the first corpus data includes speech data from multiple fields; Performing model training based on the first corpus data to obtain a second speech recognition model; Verifying the second speech recognition model to obtain a ratio of missed and false positives for target domain words; adjusting the second speech recognition model according to the missed detection and false detection ratio to obtain a first speech recognition model; The adjusting the second speech recognition model according to the missed detection to false detection ratio includes: When the ratio of missed detection to false detection is greater than 1, increasing the weight coefficient of the sample loss from the target domain in the loss function; When the ratio of missed detection to false detection is less than 1, the weight coefficient of the sample loss from the target field in the loss function is reduced, and the loss function is the weighted sum of the sample losses in multiple fields.

2. The method according to claim 1, characterized in that The missed detection of the target domain words includes: the number of times the target domain words are recognized as words from other fields, and the number of times the target domain words are not recognized; The misidentification of the target domain words includes: the number of times other domain words are recognized as the target domain words, and the number of times they are redundantly recognized as the target domain words.

3. The method according to claim 1, characterized in that The verifying of the second speech recognition model to obtain the ratio of missed detection to false detection of target domain words includes: Performing verification and recognition on the second speech recognition model using verification data to obtain a recognition result, wherein the verification data is part of the first corpus data; According to the recognition results, the ratio of missed detection to false detection of words in the target domain is obtained.

4. The method according to claim 3, characterized in that The method of obtaining the ratio of missed searches to false searches of target domain words based on the recognition results includes: Counting the recognition results to obtain missed and misidentified words in the target domain; According to the missed searches and mis-searches of the target domain words, a missed search and mis-search ratio of the target domain words is calculated.

5. The method according to claim 1, wherein Before verifying the second speech recognition model and obtaining the ratio of missed detection to false detection of target domain words, the method further includes: Performing word segmentation on the text data corresponding to the first corpus data to obtain a plurality of words; The multiple words are screened based on the fields to which they belong to obtain field words corresponding to the respective fields.

6. A speech recognition device, characterized in that: include: A first acquisition module is used to acquire voice data to be recognized, wherein the voice data to be recognized includes voice data in multiple fields; A speech recognition module, configured to perform speech recognition on the speech data to be recognized based on a pre-built first speech recognition model; The speech recognition device further comprises: A second acquisition module is used to acquire first corpus data, where the first corpus data includes speech data in multiple fields; A model training module, configured to perform model training based on the first corpus data to obtain a second speech recognition model; A model verification module, configured to verify the second speech recognition model and obtain a ratio of missed and false positives of target domain words; a first processing module, configured to adjust the second speech recognition model according to the missed detection to false detection ratio to obtain a first speech recognition model; The first processing module includes: A second processing unit is configured to increase a weight coefficient of sample loss from the target domain in a loss function when the ratio of missed detection to false detection is greater than 1; The third processing unit is used to reduce the weight coefficient of the sample loss from the target field in the loss function when the missed detection and false detection ratio is less than 1, and the loss function is the weighted loss of samples from multiple fields.

7. A terminal comprising a processor and a transceiver, wherein the transceiver receives and sends data under the control of the processor, characterized in that: The processor is configured to perform the following operations: Acquiring voice data to be recognized, wherein the voice data to be recognized includes voice data in multiple fields; Performing speech recognition on the speech data to be recognized based on a pre-built first speech recognition model; The processor is further configured to: Acquire first corpus data, where the first corpus data includes speech data from multiple fields; Performing model training based on the first corpus data to obtain a second speech recognition model; The second speech recognition model is verified to obtain the ratio of missed and false positives of target domain words. adjusting the second speech recognition model according to the missed detection and false detection ratio to obtain a first speech recognition model; The processor is further configured to: When the ratio of missed detection to false detection is greater than 1, increasing the weight coefficient of the sample loss from the target domain in the loss function; When the ratio of missed detection to false detection is less than 1, the weight coefficient of the sample loss from the target field in the loss function is reduced, and the loss function is the weighted sum of the sample losses in multiple fields. The terminal according to claim 7 , wherein: The missed detection of the target domain words includes: the number of times the target domain words are recognized as words from other fields, and the number of times the target domain words are not recognized; The misidentification of the target domain words includes: the number of times other domain words are recognized as the target domain words, and the number of times they are redundantly recognized as the target domain words.

9. The terminal according to claim 7, characterized in that The processor is further configured to: Performing verification and recognition on the second speech recognition model using verification data to obtain a recognition result, wherein the verification data is part of the first corpus data; According to the recognition results, the ratio of missed detection to false detection of words in the target domain is obtained.

10. The terminal according to claim 9, characterized in that The processor is further configured to: Counting the recognition results to obtain missed and misidentified words in the target domain; According to the missed searches and mis-searches of the target domain words, a missed search and mis-search ratio of the target domain words is calculated. The terminal according to claim 7 , wherein: The processor is further configured to: Performing word segmentation on the text data corresponding to the first corpus data to obtain a plurality of words; The multiple words are screened based on the fields to which they belong to obtain field words corresponding to the respective fields.

12. A terminal comprising a memory, a processor, and a program stored in the memory and executable on the processor; characterized in that: When the processor executes the program, the speech recognition method according to any one of claims 1 to 5 is implemented.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the speech recognition method according to any one of claims 1 to 5 are implemented.

Images